# Company evidence and current claim checks


The selected corpus establishes paying adoption, supplier revenue and some quantified operational benefits. It does not establish a common-period, fully loaded incremental end-customer AI P&L or a reconciled independent-customer-to-named-datacenter capital-return chain. Undisclosed profit remains unknown, not zero.


## Microsoft Corporation


The sale is real; the final payer is not yet reconciled.

Azure consumption and Copilot subscriptions turn useful customer work into Microsoft revenue.

Reported adoption and related-party revenue establish commercial traction. Short-lived equipment and margin pressure expose the capital-recovery clock.

Limit: Cloud includes non-AI; OpenAI revenue is not a reconciliation of independent customer cash.

Actual AI profit: undisclosed

- Management reported more than 30 million paid M365 Copilot seats and 43% Azure growth in FY2026 Q4; additional in-quarter capacity was quickly monetized. (FY2026 Q4 ended 2026-06-30).

  Limitation: Transcript paragraphs 71,149,156; Azure includes non-AI. Seats do not measure realized net price, renewal, use, or incremental profit.

  Source: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx; DOCX word/document.xml paragraphs indexed from zero: 23 throughput; 71 paid Copilot seats; 137–139 capex and cash; 143 RPO; 156 Azure; 164–165 lease reclassification; 246–260 Copilot Q&A; 275–284 investment-return Q&A

- FY2026 Microsoft Cloud gross margin fell to 66%, with AI infrastructure and product usage among the drivers, partly offset by efficiency. (FY2026 ended 2026-06-30).

  Limitation: Item 7 Fiscal Year 2026 Compared with Fiscal Year 2025, Gross margin. Global mixed cloud population; not AI-only margin.

  Source: https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm; Part I Item 1 Business; Part II Item 7 FY2026 gross margin, liquidity and Contractual Obligations table; Item 8 consolidated cash-flow statement FY2026/FY2025 columns; Note 1 Investments/OpenAI; Note 13 Leases, maturity table and uncommenced paragraph; Item 7 Cash Flows (printed p43), other-investing year-over-year change; Note 1 Contract Balances and Other Receivables (p57), component-facilitation receivables and restricted supplier investments; Note 13 supplemental cash flow (p77), FY2026 Financing cash flows from finance leases.

- Cash PP&E rose $64.551B to $115.948B; operating cash flow rose $136.162B to $182.935B between FY2025 and FY2026. (FY2025 and FY2026 ended June30).

  Limitation: Cash-flow statement columns checked in retained original HTML; worldwide consolidated, not AI-only.

  Source: https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm; Part I Item 1 Business; Part II Item 7 FY2026 gross margin, liquidity and Contractual Obligations table; Item 8 consolidated cash-flow statement FY2026/FY2025 columns; Note 1 Investments/OpenAI; Note 13 Leases, maturity table and uncommenced paragraph; Item 7 Cash Flows (printed p43), other-investing year-over-year change; Note 1 Contract Balances and Other Receivables (p57), component-facilitation receivables and restricted supplier investments; Note 13 supplemental cash flow (p77), FY2026 Financing cash flows from finance leases.

- Management described roughly two-thirds of Q4 capex as short-lived assets, principally CPUs and GPUs. (FY2026 Q4 ended 2026-06-30).

  Limitation: Transcript paragraph 137; cash-plus-finance-lease capex measure, includes AI and non-AI; not applied to annual cash PP&E.

  Source: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx; DOCX word/document.xml paragraphs indexed from zero: 23 throughput; 71 paid Copilot seats; 137–139 capex and cash; 143 RPO; 156 Azure; 164–165 lease reclassification; 246–260 Copilot Q&A; 275–284 investment-return Q&A

- FY2026 revenue from OpenAI arrangements including revenue sharing was $24.1B, with $6.0B receivable at June-end; OpenAI is an equity-method investee. (FY2026 revenue; receivable at 2026-06-30).

  Limitation: Note 1 Investments/OpenAI ASC850 disclosure; annual revenue and point-in-time receivable are different measures; not netted against investment.

  Source: https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm; Part I Item 1 Business; Part II Item 7 FY2026 gross margin, liquidity and Contractual Obligations table; Item 8 consolidated cash-flow statement FY2026/FY2025 columns; Note 1 Investments/OpenAI; Note 13 Leases, maturity table and uncommenced paragraph; Item 7 Cash Flows (printed p43), other-investing year-over-year change; Note 1 Contract Balances and Other Receivables (p57), component-facilitation receivables and restricted supplier investments; Note 13 supplemental cash flow (p77), FY2026 Financing cash flows from finance leases.

- Management said commercial RPO increased 25% excluding OpenAI and sequential RPO growth came entirely from non-frontier-model customers. (FY2026 Q4 management commentary, July2026).

  Limitation: Transcript paragraph 143; broad contracted business includes non-AI and future delivery, not collected cash or guaranteed margin.

  Source: https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/TranscriptFY26Q4.docx; DOCX word/document.xml paragraphs indexed from zero: 23 throughput; 71 paid Copilot seats; 137–139 capex and cash; 143 RPO; 156 Azure; 164–165 lease reclassification; 246–260 Copilot Q&A; 275–284 investment-return Q&A


## Meta Platforms, Inc.


Advertising AI can work without an external frontier-token sale.

Better ranking and recommendations can raise advertiser conversions, bids and retention.

Management reports concrete workflow improvements; the advertising auction is an identifiable collection mechanism.

Limit: Selected advertiser benefits do not establish typical merchant net profit or returns on the full frontier investment.

Actual AI profit: undisclosed

- Management reports Underneat saw 13% incremental purchase lift after combining Advantage+ tools. (Q2 2026 call, July2026; customer trial dates undisclosed).

  Limitation: Call p.7 selected customer example; trial population/design, merchant margin, campaign cost and persistence unavailable. Cannot infer representative ROI.

  Source: https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf; July 29, 2026; prepared remarks pp.1–9 and Q&A pp.10–21; specific model outcomes p.7, Instagram engagement p.5, open models pp.17–20, near-term capacity/2028 flexibility p.21.

- Li attributed 15.7% Facebook conversion uplift to combined user-understanding/GEM/sequence changes and double-digit Instagram time-spent growth largely to recommendations. (Q2 2026 call, July2026; uplift measurement period not specified).

  Limitation: Prepared remarks pp.5,7. Facebook uplift scope is not specified as whole-franchise revenue; experimental denominators, costs, incrementality and persistence not provided. Instagram engagement is aggregate management attribution.

  Source: https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf; July 29, 2026; prepared remarks pp.1–9 and Q&A pp.10–21; specific model outcomes p.7, Instagram engagement p.5, open models pp.17–20, near-term capacity/2028 flexibility p.21.

- $75bn Advantage+ annual revenue run rate reflects ads using tools, not a disclosed additional AI revenue stream. (Run rate reported July2026; not annual realized revenue).

  Limitation: Prepared remarks p.7; no annual realized or incremental characterization. Migration/substitution prevents addition to total advertising revenue.

  Source: https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf; July 29, 2026; prepared remarks pp.1–9 and Q&A pp.10–21; specific model outcomes p.7, Instagram engagement p.5, open models pp.17–20, near-term capacity/2028 flexibility p.21.

- More than one million businesses used agents; Li described prospective H2 charging and token-based messages from August 1. (Usage reported July2026; prospective H2 2026 charging).

  Limitation: Main p.3; follow-up pp.6,10. Adoption is not paid adoption. July forward-looking plan only; no assertion of subsequent implementation or receipts.

  Source: https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf; July 29, 2026; prepared remarks pp.1–9 and Q&A pp.10–21; specific model outcomes p.7, Instagram engagement p.5, open models pp.17–20, near-term capacity/2028 flexibility p.21.

- Substantial compute trains models; direct compute monetization had no definitive timeline; quoted premium offers are not actual realized profit. (July2026 main and follow-up calls).

  Limitation: Main Q&A p.10; follow-up p.5. Offer premiums lack full delivery/capital-cost denominator and are uncontracted management evidence.

  Source: https://s21.q4cdn.com/399680738/files/doc_financials/2026/q2/META-Q2-2026-Earnings-Call-Transcript.pdf; July 29, 2026; prepared remarks pp.1–9 and Q&A pp.10–21; specific model outcomes p.7, Instagram engagement p.5, open models pp.17–20, near-term capacity/2028 flexibility p.21.

- Operating cash grew $102.299bn to $130.301bn while cash PP&E grew $52.162bn to $89.325bn; selected residual fell $50.137bn to $40.976bn. (Four quarters July2024–June2025 versus July2025–June2026).

  Limitation: Two exact consecutive four-quarter periods: 2024-07-01–2025-06-30 and 2025-07-01–2026-06-30. Global consolidated, not AI-only or unlevered FCF.

  Source: https://www.sec.gov/Archives/edgar/data/1326801/000162828026050705/meta-20260630.htm; Original consolidated cash-flow statements and comparative columns. Q226: Notes 4,5,6,9,13 and MD&A/liquidity; K25: Note 2 advertising revenue, AI strategy and accounting-life/investment notes. Other filings: accepted financial-panel cash operands; not a fresh full narrative reread.


## Amazon


Growing cloud profit and delayed billing can coexist.

AWS sells useful compute with attached services and custom silicon, while retail can capture AI benefits internally.

AWS has reported revenue and operating profit; management says capital commonly precedes billing by 6–24 months.

Limit: Segment profit is mixed AI/non-AI; run rates overlap and are not collected annual revenue. Timing is management guidance.

Actual AI profit: undisclosed

- Q2 2026 AWS sales $42.232bn, +37%; operating income $16.621bn; Q2 2025 sales $30.873bn and operating income $10.160bn. (Q2 2026 versus Q2 2025).

  Limitation: Global AWS, including non-AI services.

  Source: https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm; Cash-flow statement, H1 columns; Note 1 Unearned Revenue; Note 2 Anthropic/OpenAI and liquid/restricted assets; Notes 3–5 lease, commitment and financing disclosures; MD&A AWS sales and operating income.

- Management says AI annual revenue run rate exceeds $25bn and Q2 Bedrock spend exceeds all earlier quarters combined. (Q2 2026 management report; run rate not annual revenue).

  Limitation: Run rate not realized annual revenue; no AI profit or average customer ROI established.

  Source: https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Second-Quarter-Results/; Dated July 30 2026; Jassy quotation and highlights: AI/chips run rates, Bedrock spending, Trainium adoption; rendered browser snapshot.

- Chip business run rate exceeds $25bn; includes non-AI chips and AWS monetization overlapping AI services. (Management run rate reported in 2026 shareholder letter).

  Limitation: Letter defines Graviton/Trainium/Nitro and monetization through EC2. Do not sum service/chip run rates.

  Source: https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Second-Quarter-Results/; Dated July 30 2026; Jassy quotation and highlights: AI/chips run rates, Bedrock spending, Trainium adoption; rendered browser snapshot.

- Subset of server/network lives cut six to five years January 2025, adding $1.4bn 2025 depreciation primarily AWS. (Effective January2025; depreciation effect FY2025).

  Limitation: Accounting estimate does not establish economic obsolescence of every AI asset.

  Source: https://www.sec.gov/Archives/edgar/data/1018724/000101872426000004/amzn-20251231.htm; Note 1: January 2025 useful-life reduction; Business: Developers and Enterprises; segment infrastructure allocation.

- Management says capex commonly precedes billing 6–24 months depending on component; much 2026 AWS capex monetized in 2027–2028. (2026 shareholder letter; expected 2027–2028 monetization).

  Limitation: Expected timing and attractive cumulative returns are management forecasts, not realized cohort evidence.

  Source: https://s2.q4cdn.com/299287126/files/doc_financials/2026/ar/2025-Shareholder-Letter-Final.pdf; AI numbered points 2–6: customer platform rationale, Trainium price-performance, chip run-rate composition, expected margin/capex savings, billing lag and 2027–2028 monetization. Date supported by retained IR index.

- OpenAI $100bn expansion over eight years; Anthropic more than $100bn expansion over ten years; chip-performance obligations qualify commitments. (Agreements reported in Q2 2026 filing; eight/ten-year prospective terms).

  Limitation: Long-term commitments not already earned revenue or independent end-customer cash. Amounts not added to investments/capex.

  Source: https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm; Cash-flow statement, H1 columns; Note 1 Unearned Revenue; Note 2 Anthropic/OpenAI and liquid/restricted assets; Notes 3–5 lease, commitment and financing disclosures; MD&A AWS sales and operating income.


## Alphabet Inc.


A stronger ad and cloud business does not identify the AI increment.

Search ads monetize user attention; Cloud sells AI and non-AI services and hardware.

Management reports improved conversions and lower response costs alongside filed segment growth.

Limit: AI adoption, Wiz, TPU sales and non-AI migration coexist; all Cloud operating profit cannot be assigned to AI infrastructure.

Actual AI profit: undisclosed

- Q2 Search & other revenue increased from 54.190B to 63.271B USD. (Q2 2026 versus Q2 2025).

  Limitation: Note 2 three-month 2025/2026 columns, not AI-only.

  Source: https://www.sec.gov/Archives/edgar/data/1652044/000165204426000071/goog-20260630.htm; Balance sheet p4; cash flows p9 six-month 2026 column; Notes 1–4 pp10–24, Note 6 debt, Note 8 Wiz, Note 10 commitments, Note 11 equity, Note 15 segments; MD&A liquidity pp51–54.

- AI Max/PMax adopters average 15% more Search conversions or value at similar ROAS, per management. (Q2 2026 call; adopter measurement period not specified).

  Limitation: Schindler prepared remarks; selected adopters, not controlled causal profit attribution.

  Source: https://abc.xyz/investor/events/event-details/2026/2026-Q2-Earnings-Call-2026-GgTAq7Is0z/default.aspx; Schindler Search/adopter conversions; Pichai paid enterprise use; CFO results/outlook; Anmuth financing Q&A; Sheridan TPU inventory/revenue; Nathanson product margin; Shmulik bridge-capacity return.

- Management reports >30% core AI response cost decline since Gemini 3 upgrade in April and further AI Mode improvements in July. (Since April2026 Gemini3 upgrade; July2026 commentary).

  Limitation: No comparable dollar cost or query contribution disclosed; usage and task mix matter.

  Source: https://abc.xyz/investor/events/event-details/2026/2026-Q1-Earnings-Call-2026-nW8kCrBAKS/default.aspx; Pichai Search response costs; Schindler monetization; CFO Cloud and TPU outlook; Q&A.

- Q2 Cloud revenue 24.768B and operating profit 8.814B, 82% revenue growth and margins 20.7% to 35.6%. (Q2 2026 versus Q2 2025).

  Limitation: Includes non-AI, Wiz and new TPU sales; shared research partly outside segment.

  Source: https://www.sec.gov/Archives/edgar/data/1652044/000165204426000071/goog-20260630.htm; Balance sheet p4; cash flows p9 six-month 2026 column; Notes 1–4 pp10–24, Note 6 debt, Note 8 Wiz, Note 10 commitments, Note 11 equity, Note 15 segments; MD&A liquidity pp51–54.

- Frontier-model development is baseline compute allocation; shared AI research partly recorded centrally. (None).

  Limitation: Do not charge all central research solely to Cloud.

  Source: https://abc.xyz/investor/events/event-details/2026/2026-Q2-Earnings-Call-2026-GgTAq7Is0z/default.aspx; Schindler Search/adopter conversions; Pichai paid enterprise use; CFO results/outlook; Anmuth financing Q&A; Sheridan TPU inventory/revenue; Nathanson product margin; Shmulik bridge-capacity return.


## CoreWeave, Inc.


A contract can move utilization risk; it cannot abolish renewal risk.

Committed GPU capacity pays the operator before the asset becomes uneconomic, provided delivery and customer payment occur.

Committed revenue and RPO are real business anchors; finite contract duration makes first-term capital recovery decisive.

Limit: RPO is not cash or profit; six-year accounting life is not verified economic life. The A100 renewal claim is evidence against universal rapid obsolescence.

Actual AI profit: undisclosed

- Microsoft supplied 67% of 2025 revenue; OpenAI and Meta have material commitments. (FY2025 revenue concentration).

  Limitation: 2025 concentration; do not map anonymized Q2 letters across periods.

  Source: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm; Business/platform, customer concentration and competitors; MD&A Components of Results; Note1 technology-equipment life and committed-contract revenue policy

- 98% of Q2 revenue came from commitments; take-or-pay shifts immediate utilization risk but leaves delivery/credit/renewal risks. (Q2 2026).

  Limitation: 98% is reported; implication conditional on contract delivery/credit. Public summaries do not expose all legal terms.

  Source: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm; Cash flows p11 (six months 2026 column); balance sheet p6; Note2 p15; Note8 pp24–25; Note10 pp27–31; MD&A liquidity and customer/supplier risks

- June30 RPO was $103.7bn, undelivered transaction price adjusted for specified variable consideration. (2026-06-30 balance).

  Limitation: RPO is neither cash nor profit; includes billed/unbilled. No backlog-times-margin calculation.

  Source: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm; Cash flows p11 (six months 2026 column); balance sheet p6; Note2 p15; Note8 pp24–25; Note10 pp27–31; MD&A liquidity and customer/supplier risks

- Management says typical five-year deployment pays asset debt plus substantial cash, earning return before renewal. (August2026 management description of typical five-year deployment).

  Limitation: No disclosed cohort waterfall proving all-capital return; debt payoff alone insufficient.

  Source: https://s205.q4cdn.com/133937190/files/doc_financials/2026/q2/CRWV-US-CORRECTED-TRANSCRIPT-CoreWeave-Q2-2026-Earnings-Call-11August2026.pdf; pp3,7–9,10–13,16–18: Intrator/Agrawal prepared remarks; Chatterjee renewal/pricing, Zelnick inference/margins, Turrin power timing, Knoblauch duration Q&A

- A100 contract through2029 and July approximate25% price rise; Q2 new-contract expected contribution margins +5–10ppt. (August2026 commentary; A100 contract through2029; July2026 pricing).

  Limitation: Architecture launch2020 does not show age of installed machine. Expected contribution is not realized after-tax cash margin; price includes supplier pass-through.

  Source: https://s205.q4cdn.com/133937190/files/doc_financials/2026/q2/CRWV-US-CORRECTED-TRANSCRIPT-CoreWeave-Q2-2026-Earnings-Call-11August2026.pdf; pp3,7–9,10–13,16–18: Intrator/Agrawal prepared remarks; Chatterjee renewal/pricing, Zelnick inference/margins, Turrin power timing, Knoblauch duration Q&A

- Renewal-eligible fleet was very limited; technology equipment accounting life is six years. (August2026 Q2 call and June2026 filing accounting policy).

  Limitation: Fleet size statement is management Q&A; six-year estimate does not ensure economic life.

  Source: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm; Business/platform, customer concentration and competitors; MD&A Components of Results; Note1 technology-equipment life and committed-contract revenue policy

- Two/three-year enterprise contracts became financeable; September three/six-month contracts quoted annualized$40m/MW. (August/September2026 commentary; September contracts last three/six months).

  Limitation: Annualized rate not guaranteed year revenue; no volume, capex or useful-output denominator supplied.

  Source: https://s205.q4cdn.com/133937190/files/doc_financials/2026/q2/CRWV-US-CORRECTED-TRANSCRIPT-CoreWeave-Q2-2026-Earnings-Call-11August2026.pdf; pp3,7–9,10–13,16–18: Intrator/Agrawal prepared remarks; Chatterjee renewal/pricing, Zelnick inference/margins, Turrin power timing, Knoblauch duration Q&A

- H1 CFO$3.663bn, cashPPE$14.117bn, net investing$14.874bn and net financing$13.985bn imply$11.211bn prefinancing deficit. (Six months ended 2026-06-30).

  Limitation: Consolidated global cash flows; includes expansion, not a cohort return or AI-only profit.

  Source: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm; Cash flows p11 (six months 2026 column); balance sheet p6; Note2 p15; Note8 pp24–25; Note10 pp27–31; MD&A liquidity and customer/supplier risks


## Oracle Corporation


Cash paid in advance is genuine financing, not earned profit.

Cloud contracts and customer hardware contributions can fund delivery before Oracle recognizes revenue.

The filing identifies substantial customer prepayments and an expanding infrastructure business.

Limit: Do not mistake a $75bn hardware contract portion for cash receipts, or RPO for earned revenue; blended margin is not AI margin.

Actual AI profit: undisclosed

- FY27 Q1 cloud infrastructure $7.388bn grew121%; software support $4.895bn approximately flat. (FY2027 Q1 ended 2026-08-31).

  Limitation: Worldwide offerings, neither AI-only nor separately profited.

  Source: https://www.sec.gov/Archives/edgar/data/1341439/000119312526389274/orcl-20260831.htm; Cash flows p.5; Note1 pp.6–7; Note2 p.9; Note3 pp.9–10; Note6 pp.12–13; MD&A pp.27–28,33–35

- Most Q3/Q4 RPO growth involved prepaid/customer-supplied GPUs; hardware portions $75bn. (FY2026 Q3/Q4 RPO growth; reported June2026).

  Limitation: Contract portion, not $75bn receipts. Contract ownership, price/replacement terms not disclosed.

  Source: https://s23.q4cdn.com/440135859/files/doc_earnings/2026/q4/earnings-result/4q26-pressrelease-final.pdf; pp.2–3, RPO and capital funding

- FY27Q1 CFO$23.103bn includes$11.363bn significant-financing prepayments; interest accounted separately. (FY2027 Q1 ended 2026-08-31).

  Limitation: Pre-service collections not earned profit; no second cash inflow on recognition.

  Source: https://www.sec.gov/Archives/edgar/data/1341439/000119312526389274/orcl-20260831.htm; Cash flows p.5; Note1 pp.6–7; Note2 p.9; Note3 pp.9–10; Note6 pp.12–13; MD&A pp.27–28,33–35

- Reported CFO-minus-capex gap$5.396bn; excluding identified advances gap$16.759bn. (FY2027 Q1 ended 2026-08-31).

  Limitation: Not normalized cash, no AI attribution.

  Source: https://www.sec.gov/Archives/edgar/data/1341439/000119312526389274/orcl-20260831.htm; Cash flows p.5; Note1 pp.6–7; Note2 p.9; Note3 pp.9–10; Note6 pp.12–13; MD&A pp.27–28,33–35

- 664bn RPO, approximately13% expected revenue next12months; availability/usage govern recognition. (2026-08-31 balance; prospective recognition).

  Limitation: Expected conversion is management estimate; not receipts/profit or contract-by-contract conditions.

  Source: https://www.sec.gov/Archives/edgar/data/1341439/000119312526389274/orcl-20260831.htm; Cash flows p.5; Note1 pp.6–7; Note2 p.9; Note3 pp.9–10; Note6 pp.12–13; MD&A pp.27–28,33–35

- Six-year server/network accounting life;288bn additional uncommenced leases,15–19years, start Q2FY27–FY29. (2026-08-31; prospective lease commencements FY2027Q2–FY2029).

  Limitation: Lease face value not loss/penalty; economic chip life may differ.

  Source: https://www.sec.gov/Archives/edgar/data/1341439/000119312526389274/orcl-20260831.htm; Cash flows p.5; Note1 pp.6–7; Note2 p.9; Note3 pp.9–10; Note6 pp.12–13; MD&A pp.27–28,33–35

- Cloud/software segment margin60%→55% despite greater margin dollars. (FY2027 Q1 versus FY2026 Q1).

  Limitation: Blended and excludes stock compensation/certain allocations/intangible amortization/other GAAP costs; not AI margin.

  Source: https://www.sec.gov/Archives/edgar/data/1341439/000119312526389274/orcl-20260831.htm; Cash flows p.5; Note1 pp.6–7; Note2 p.9; Note3 pp.9–10; Note6 pp.12–13; MD&A pp.27–28,33–35


## Equinix, Inc.


Useful power and connectivity have value; cabinet counts are not AI demand.

Customers pay for colocated infrastructure, power access and interconnection.

Existing-customer bookings and billed capacity support a real established franchise.

Limit: The 27% mature-site gross-profit yield is not a project IRR; cabinets, MW, presales and powered service are different quantities.

Actual AI profit: undisclosed

- Q2 2026 normalized constant-currency MRR growth 11%; 4,200 billed cabinets and 9,700 interconnections added; MRR/cab rose 6% normalized at constant currency. (Q2 2026).

  Limitation: Company adjusted metrics and selected non-financial perimeter; not AI-only growth. MRR/cab includes density and mix, not pure repricing.

  Source: https://d1io3yog0oux5.cloudfront.net/_f9ad8b88d669de65a966b7e00da37b99/equinix/db/2183/27054/earnings_presentation/Equinix+Q2+26+Earnings+Presentation+Final.pdf; Printed pp.3,10,11,24,26,48–50: demand, mature portfolio, capex, cabinet metrics, construction schedule and definitions

- More than 90% of recurring bookings in preceding three years came from existing customers; initial contracts generally one to five years. (Preceding three years, reported in FY2025 annual filing).

  Limitation: Global company recurrence does not establish AI demand incrementality or twenty-year contractual cash protection.

  Source: https://www.sec.gov/Archives/edgar/data/1101239/000110123926000147/eqix-20260630.htm; pp.6/9 balance sheet and cash flows; Note4 pp.12–14; Note7 leases; Note9 commitments; MD&A revenue, AFFO pp.55–56 and liquidity

- Management mature-site 27% ratio is trailing cash gross profit over historical gross PP&E; displayed $5,232m / $19,639m rounds to 27%. (Q2 2026 presentation, trailing mature-site cash-gross-profit metric).

  Limitation: Constant-currency ratio; selected mature portfolio; excludes xScale JVs/Metal/non-IBX tenant income/non-data-centre assets. Not project IRR or net equity return.

  Source: https://d1io3yog0oux5.cloudfront.net/_f9ad8b88d669de65a966b7e00da37b99/equinix/db/2183/27054/earnings_presentation/Equinix+Q2+26+Earnings+Presentation+Final.pdf; Printed pp.3,10,11,24,26,48–50: demand, mature portfolio, capex, cabinet metrics, construction schedule and definitions

- Gross-profit yield omits cash SG&A, tax, financing and capital maintenance; includes leased-property economics and historical cost, not replacement cost. (None).

  Limitation: Deck definitions of cash gross profit, leased mature sites and cost base establish boundary; inference is that the ratio does not establish marginal fully costed return.

  Source: https://d1io3yog0oux5.cloudfront.net/_f9ad8b88d669de65a966b7e00da37b99/equinix/db/2183/27054/earnings_presentation/Equinix+Q2+26+Earnings+Presentation+Final.pdf; Printed pp.3,10,11,24,26,48–50: demand, mature portfolio, capex, cabinet metrics, construction schedule and definitions

- Mature-site utilization 82%, worldwide cabinet utilization 78%, roughly 30% of 2026 expansion cabinets presold. (Q2 2026 presentation; presales apply 2026 expansion).

  Limitation: Deck pp.10/24/26 use distinct populations. Cabinets are not AI MW and presold is not utility-delivered.

  Source: https://d1io3yog0oux5.cloudfront.net/_f9ad8b88d669de65a966b7e00da37b99/equinix/db/2183/27054/earnings_presentation/Equinix+Q2+26+Earnings+Presentation+Final.pdf; Printed pp.3,10,11,24,26,48–50: demand, mature portfolio, capex, cabinet metrics, construction schedule and definitions

- First-half CFO $1,784m versus $1,753m and PP&E purchases $2,834m versus $1,739m in 2026 versus 2025; dividends $1,029m in 2026. (Six months ended June30,2026 versus2025).

  Limitation: Global consolidated, first-half periods, purchases of other PP&E exclude real estate acquisitions; CFO includes interest and working capital.

  Source: https://www.sec.gov/Archives/edgar/data/1101239/000110123926000147/eqix-20260630.htm; pp.6/9 balance sheet and cash flows; Note4 pp.12–14; Note7 leases; Note9 commitments; MD&A revenue, AFFO pp.55–56 and liquidity


## OpenAI


Paying accounts establish a market, not its lifetime margin.

Consumer subscriptions, business seats and API services monetize completed useful work.

Company-reported paying populations and counterparties’ filed commercial balances establish substantial activity.

Limit: Counts may overlap; prices, net collections, retention and inference intensity are not disclosed on one common cohort. Financing is not a loss or evidence of fraud.

Actual AI profit: undisclosed

- OpenAI reported >50m consumer subscribers, >9m paying business users and >900m weekly active ChatGPT users in February2026. (2026-02-27 announcement snapshot).

  Limitation: Self-reported counts, not recognized revenue, retention, separate unique populations or demonstrated savings; no ratio is computed across them.

  Source: https://openai.com/index/scaling-ai-for-everyone/; Article paragraphs beginning Today, More than 9 million, and ChatGPT is where; funding, users and subscribers

- April2026 terms permit OpenAI all-product cross-cloud serving; Microsoft keeps nonexclusive model/product IP through2032, stops paying revenue share, and receives OpenAI revenue share through2030 at unchanged undisclosed percentage with a cap. (April2026 amended relationship summary).

  Limitation: Official summary, not executed contract. Cap/percentage/schedule and actual expense undisclosed. April amendment supersedes relevant October exclusivity/revenue-share statements.

  Source: https://openai.com/index/next-phase-of-microsoft-partnership/; Agreement bullet list: cloud access, IP licence, two revenue-share directions and cap

- Microsoft recorded$24.1bn FY2026 commercial/revenue-share revenue involving OpenAI, $6bn June30 receivable and$11.9bn funded of$13bn investment commitment. (Microsoft FY2026; receivable/investment funded at June30,2026).

  Limitation: Microsoft consolidated counterparty accounting. Not OpenAI audited revenue, expense, profit or cash. Annual flow, point-in-time receivable and cumulative investment cannot be added or netted; no default allegation.

  Source: https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm; Note1 Investment in OpenAI/related-party paragraph; commercial revenue, receivable, funding, HLBV; Note3 recapitalization gains

- Amazon reported$28.7bn invested byJune30 and remaining$21.3bn subsequently, establishing$50bn funded by the July31filing. (Amazon cash through June30,2026 plus subsequent event before July31 filing).

  Limitation: Subsequent event is after quarter-end and before filing, not June30 cash. Does not establish OpenAI remaining unrestricted cash or funding from other investors.

  Source: https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm; Note1 Unearned Revenue AWS/OpenAI $38bn+$100bn/8years and chip-performance obligations; Note2 Equity Investments OpenAI SeriesC; MD&A Liquidity printed pages27–28

- The AWS commitment grew$100bn over8years from existing$38bn and includes AWS-chip-performance obligations. (Eight-year commitment expansion reported in2026).

  Limitation: This is a commercial commitment, not period expense or supplier capex. Public note does not specify complete payment/cancellation terms or justify straight-line annual expense.

  Source: https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm; Note1 Unearned Revenue AWS/OpenAI $38bn+$100bn/8years and chip-performance obligations; Note2 Equity Investments OpenAI SeriesC; MD&A Liquidity printed pages27–28

- Supplier equity can bridge rational investment, but supplier sales financed by investors cannot alone establish independent end-customer economics; further equity dilutes earlier owners. (None).

  Limitation: No unsupported fraud/circular-revenue allegation, dollar-for-dollar offset, loss claim or statement that new equity necessarily denotes failure.

  Source: https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm; Note1 Investment in OpenAI/related-party paragraph; commercial revenue, receivable, funding, HLBV; Note3 recapitalization gains


## Palantir Technologies Inc.


Application value can grow faster than the upstream model bill.

An integrated workflow can earn subscription and service fees while routing tasks among third-party and customer-owned models.

Reported commercial growth and margin support application monetization; pilots can be free and commitments have distinct definitions.

Limit: None of the disclosed totals isolates AIP profit or end-customer ROI; the falling-model-bill example is an assumption, not actual Palantir telemetry.

Actual AI profit: undisclosed

- Q2 2026 consolidated revenue $1,935.464m, GAAP operating income $912.004m, and U.S. commercial revenue $764m (149% YoY growth). (Q2 2026).

  Limitation: Worldwide issuer totals and U.S. commercial product mix are not AIP-only.

  Source: https://www.sec.gov/Archives/edgar/data/1321655/000132165526000041/pltr-20260630.htm; Part I Item 1 Notes 3 (RPO), 7 (Purchase Commitments); Part I Item 2 Overview/Revenue/Customers/Contribution Margin/Non-GAAP Reconciliations; condensed operations, cash flows and June balance-sheet tables

- Palantir often conducts pilots and bootcamps at its own expense or no/low customer cost and warns of minimal future revenue. (Business practice described in June2026 filing).

  Limitation: No disclosed bootcamp conversion, production-task cohort or renewal rate.

  Source: https://www.sec.gov/Archives/edgar/data/1321655/000132165526000041/pltr-20260630.htm; Part I Item 1 Notes 3 (RPO), 7 (Purchase Commitments); Part I Item 2 Overview/Revenue/Customers/Contribution Margin/Non-GAAP Reconciliations; condensed operations, cash flows and June balance-sheet tables

- Q2 2026 commercial revenue increase of $495m YoY included $407m from customers existing December 31, 2025; trailing-year revenue-producing customers numbered 1,049 versus 849. (Q2 2026 versus Q2 2025; existing customers defined at December31,2025).

  Limitation: Existing-customer expansion is not AIP-specific, and cohort definition is not a retention statistic.

  Source: https://www.sec.gov/Archives/edgar/data/1321655/000132165526000041/pltr-20260630.htm; Part I Item 1 Notes 3 (RPO), 7 (Purchase Commitments); Part I Item 2 Overview/Revenue/Customers/Contribution Margin/Non-GAAP Reconciliations; condensed operations, cash flows and June balance-sheet tables

- Q2 2026 U.S. commercial TCV $2.132bn and June U.S. commercial RDV $6.238bn; definitions assume options and no termination although many contracts can terminate. (Q2 2026 TCV; June30,2026 RDV).

  Limitation: TCV is potential lifetime value, not quarterly or annual revenue; RDV is not noncancelable RPO.

  Source: https://investors.palantir.com/news-details/2026/Palantir-Reports-Q2-2026-U-S--Comm-Revenue-Growth-of-149-YY-and-Revenue-Growth-of-93-YY-Raises-FY-2026-Revenue-Guidance-to-82-YY-Growth-and-U-S--Comm-Revenue-Guidance-to-134-YY-Crushing-Consensus-Expectations/; Rendered main article: Q2 2026 Highlights; Financial Summary; Additional Definitions for TCV/RDV/RPO; Non-GAAP Reconciliations for operating income and stock compensation

- Company-wide June 2026 RPO was $4.9bn noncancelable contracted revenue, about 43% expected in next 12 months. (June30,2026 balance; prospective next12months).

  Limitation: RPO excludes cancellable deals and contracts with original term <=12 months under practical expedient; not AIP-specific or cash.

  Source: https://www.sec.gov/Archives/edgar/data/1321655/000132165526000041/pltr-20260630.htm; Part I Item 1 Notes 3 (RPO), 7 (Purchase Commitments); Part I Item 2 Overview/Revenue/Customers/Contribution Margin/Non-GAAP Reconciliations; condensed operations, cash flows and June balance-sheet tables

- March 2026 cloud-services agreement amendment requires at least $5.6bn spend over ten contract years through Feb 2036, with annual minimums $268m-$979m; prior obligations under same agreement terminated. (March2026 amendment through February29,2036).

  Limitation: Services not identified as AIP tokens, GPUs, new incremental datacenter capex or a particular supplier sale.

  Source: https://www.sec.gov/Archives/edgar/data/1321655/000132165526000041/pltr-20260630.htm; Part I Item 1 Notes 3 (RPO), 7 (Purchase Commitments); Part I Item 2 Overview/Revenue/Customers/Contribution Margin/Non-GAAP Reconciliations; condensed operations, cash flows and June balance-sheet tables

- Palantir does not disclose a separate AIP profit, capital cohort, production-task volume, billed token intensity for external routes, realized token price/mix or externally served task share. (Selected sources available by September27,2026).

  Limitation: Bounded assessment of selected original sources; absence is not proof the values cannot be disclosed elsewhere.

  Source: https://www.sec.gov/Archives/edgar/data/1321655/000132165526000041/pltr-20260630.htm; Part I Item 1 Notes 3 (RPO), 7 (Purchase Commitments); Part I Item 2 Overview/Revenue/Customers/Contribution Margin/Non-GAAP Reconciliations; condensed operations, cash flows and June balance-sheet tables

- Illustrative doubled useful tasks can coincide with half the external model-provider bill if externally served tokens/task and realized price each halve while externally served task share remains fixed. (None).

  Limitation: Task population includes all accepted tasks; externally_billed_share removes customer-owned/self-hosted tasks once. All values are assumed for one hypothetical workflow; neither row estimates Palantir actuals.

  Source: None; Authored illustrative calculation PLTR-C14; not an observed source fact


## Apple Inc.


A useful local assistant need not sell an external model call.

AI can support device sales and retention through on-device, Private Cloud Compute and optional external routes.

Deployment architecture shows why app usefulness and paid external tokens are distinct; AI-related R&D is mentioned without allocation.

Limit: Product growth is not AI-attributed; beta promises are not deployed adoption; partner settlement and routing are undisclosed.

Actual AI profit: undisclosed

- Siri AI demo includes personal-message and email search and cross-app actions; user beta was promised later in 2026. (June8,2026 demonstration; beta planned later2026).

  Limitation: Developer demonstration and planned beta do not establish adoption or successful tasks.

  Source: https://www.apple.com/newsroom/2026/06/apple-introduces-siri-ai-a-profoundly-more-capable-and-personal-assistant/; Article dated June 8 2026; personal-context examples; on-device model and Spotlight/App Toolbox; beta availability.

- FY2026 Q3 iPhone sales $54.252bn versus $44.582bn; Services $30.739bn versus $27.423bn; management cites Pro models and advertising/cloud. (FY2026 Q3 versus FY2025 Q3).

  Limitation: Worldwide quarter; not an AI causal estimate.

  Source: https://www.sec.gov/Archives/edgar/data/320193/000032019326000020/aapl-20260627.htm; Item 1 consolidated operations/cash flow, Note 9 purchase obligations, Item 2 Products and Services Performance, Gross Margin, R&D.

- Latest Apple Foundation Models were built with Google/Gemini collaboration and run on device and Apple Private Cloud Compute; Siri local components include Spotlight/App Toolbox. (June8,2026 announcement).

  Limitation: Deployment route does not disclose the partner economic settlement.

  Source: https://www.apple.com/newsroom/2026/06/apple-intelligence-brings-powerful-ai-capabilities-into-everyday-experiences/; Article dated June 8 2026; A Bold New Architecture, Built Privacy-First; availability footnotes.

- Optional ChatGPT integration is available via Siri suggestion and Writing Tools; account not required for default access. (December11,2024 announcement).

  Limitation: Provider billings, usage, settlement, revenue share and subscription conversion undisclosed.

  Source: https://www.apple.com/newsroom/2024/12/apple-intelligence-now-features-image-playground-genmoji-and-more/; Article dated December 11 2024; Tap into ChatGPT with Siri and Writing Tools; privacy and availability.

- FY2026 Q3 R&D $11.729bn versus $8.866bn; Apple cites infrastructure costs including AI and headcount, without AI allocation. (FY2026 Q3 versus FY2025 Q3).

  Limitation: The $2.863bn year-over-year increase is not an AI-only cost.

  Source: https://www.sec.gov/Archives/edgar/data/320193/000032019326000020/aapl-20260627.htm; Item 1 consolidated operations/cash flow, Note 9 purchase obligations, Item 2 Products and Services Performance, Gross Margin, R&D.

- June 2026 unconditional purchase obligations with remaining terms over one year totaled $27.628bn, including $2.053bn due in the remaining three months of FY2026, across suppliers, IP, content and distribution rights. (June2026 balance; FY2026 remaining-quarter payments).

  Limitation: Original contract term is not the same as remaining payment term; obligations cannot be assigned to AI from the disclosed table.

  Source: https://www.sec.gov/Archives/edgar/data/320193/000032019326000020/aapl-20260627.htm; Item 1 consolidated operations/cash flow, Note 9 purchase obligations, Item 2 Products and Services Performance, Gross Margin, R&D.


## Anthropic


Revenue pace and financing answer different questions.

Model and coding services charge for customer use; financing supplies capital before retained earnings suffice.

A direct model provider reports substantial monetization.

Limit: No matched period collections, customer profitability, coding share or workload margin established by this announcement.

Actual AI profit: unknown

- Anthropic stated its revenue run rate crossed $47bn earlier in May 2026; it announced $65bn financing including $15bn previously committed hyperscaler investments. (May2026; announcement May28).

  Limitation: Run rate, financing and previously committed investment are different measures; do not add included $15bn again.

  Source: https://www.anthropic.com/news/series-h; Retained original UTF-8 lines 23–29; run rate line27, included prior commitments line29


## C.H. Robinson


Faster quotes coexist with a large non-AI earnings confounder.

Financial corroboration; causal AI share remains unisolated

Concrete operational or accounting example, with the original review scope preserved.

Limit: A fast quote is not automatically a booked load or incremental contribution profit. The full transcript and all historical filings remain to be collected. NAST improved while Global Forwarding weakened: company averages obscure different economics. Divestiture charges, perimeter changes and prior-year restructuring affect the comparison. The residual is not automatically AI. No complete source-transcript/filing corpus reviewed; do not treat this seed grade as a validated investment ranking.

Actual AI profit: unknown

- Management described freight-quote coverage rising from 60% to 100% and cycle time falling from 17–20 minutes to under 32 seconds. The book prints a repeated excerpt; it is one observation. (FY2025 vs FY2024).

  Limitation: Full original transcript not reopened; digest cannot establish net profit.

  Source: None; Orange Book PDF printed pages 45

- FY2025 operating income was $794.961m versus $669.141m; average headcount was 12,733 versus 14,386. (FY2025 vs FY2024).

  Limitation: Prior selected primary-filing reconciliation retained in seed; original bytes not located/reopened in this bounded pass.

  Source: https://investor.chrobinson.com/news/press-releases/news-details/2026/C-H--Robinson-Reports-2025-Fourth-Quarter-Results/default.aspx; Retained curated seed claims; original filing sections []

- Segment operating-income changes reconcile: NAST +$90.544m, Global Forwarding −$28.693m, Other/Corporate +$63.969m. (FY2025 vs FY2024).

  Limitation: Prior selected primary-filing reconciliation retained in seed; original bytes not located/reopened in this bounded pass.

  Source: https://www.sec.gov/Archives/edgar/data/1043277/000104327726000009/chrw-20251231.htm; Retained curated seed claims; original filing sections ['MD&A', 'Note 8 — Segment Reporting', 'Note 14 — Restructuring', 'Note 15 — Divestitures']

- Europe-divestiture expenses declined from $44.462m to $2.108m before tax. The $42.354m difference is about 33.7% of the $125.820m total operating-income increase. (FY2025 vs FY2024).

  Limitation: Prior selected primary-filing reconciliation retained in seed; original bytes not located/reopened in this bounded pass.

  Source: https://www.sec.gov/Archives/edgar/data/1043277/000104327726000009/chrw-20251231.htm; Retained curated seed claims; original filing sections ['MD&A', 'Note 8 — Segment Reporting', 'Note 14 — Restructuring', 'Note 15 — Divestitures']

- The 2025 restructuring program explicitly includes AI, automation and facilities consolidation; recognized program charges were $30.405m. (FY2025 vs FY2024).

  Limitation: Prior selected primary-filing reconciliation retained in seed; original bytes not located/reopened in this bounded pass.

  Source: https://www.sec.gov/Archives/edgar/data/1043277/000104327726000009/chrw-20251231.htm; Retained curated seed claims; original filing sections ['MD&A', 'Note 8 — Segment Reporting', 'Note 14 — Restructuring', 'Note 15 — Divestitures']


## Bank of America


Two thousand people of capacity is not two thousand payroll exits.

Coding capacity evidence; payroll conversion not established

Concrete operational or accounting example, with the original review scope preserved.

Limit: Equivalent capacity is not evidence that 2,000 paid roles were removed. The coding stage is only part of the end-to-end product-delivery process. No complete source-transcript/filing corpus reviewed; do not treat this seed grade as a validated investment ranking.

Actual AI profit: unknown

- Management connected AI coding tools to capacity equivalent to about 2,000 people and discussed allowing headcount to decline through attrition. (Q4 2025 / early-2026 statements).

  Limitation: Full original transcript not reopened; digest cannot establish net profit.

  Source: None; Orange Book PDF printed pages 15,44


## McKesson


Avoided escalations are a workflow measure, not company-wide savings.

Narrow operational KPI; dollar savings undisclosed

Concrete operational or accounting example, with the original review scope preserved.

Limit: The denominator is a specific regulatory-data inquiry workflow, not all support contacts or all operating cost. Avoided escalations can improve service without releasing payroll. No complete source-transcript/filing corpus reviewed; do not treat this seed grade as a validated investment ranking.

Actual AI profit: unknown

- An AI chat tool launched in November reportedly prevented 75% of DSCSA-related inquiries from escalating. (Q4 2025 / early-2026 statements).

  Limitation: Full original transcript not reopened; digest cannot establish net profit.

  Source: None; Orange Book PDF printed pages 30,46


## Cognizant


Contract design decides who keeps the coding gain.

Per-employee improvement; contract mix affects capture

Concrete operational or accounting example, with the original review scope preserved.

Limit: Fixed-price versus time-and-materials contracts change who keeps productivity gains. Acquisitions, utilization and geography can change per-employee metrics. No complete source-transcript/filing corpus reviewed; do not treat this seed grade as a validated investment ranking.

Actual AI profit: unknown

- Management reported 5% higher trailing revenue and 8% higher adjusted operating income per employee, alongside AI-assisted development and a higher fixed-bid/transaction-based mix. (Q4 2025 / early-2026 statements).

  Limitation: Full original transcript not reopened; digest cannot establish net profit.

  Source: None; Orange Book PDF printed pages 46


## CUR-C01: Anthropic leaked S-1 is impressive but growth is slowing.


Reported quarterly revenue still grew rapidly, while its percentage growth slowed. The draft-filing claims remain attributed reporting.

Euronews reports Q3 revenue16.8B versus Q2 11.5B:46.1% sequential growth after a more-than-doubling quarter. WSJ reports reviewing a draft prospectus. The original draft was not obtained.

Still needed: Original draft S-1, version/date and revenue footnotes; Comparable Q1–Q3 financial statements; Definition of adjusted profit and cash-flow adjustments


## CUR-C02: Public OpenRouter token growth slowed in the observed twelve-week window; the latest completed week declined.


Public OpenRouter token growth decelerated and the latest complete week dipped. This is not evidence of persistent contraction or total-market demand decline.

Twelve primary rounded weekly totals are retained for July27–October18. Equal four-week totals rose294.5T→484T→648T, slowing from+64.3% to+33.9%. Latest week164T fell approximately4.7% from172T; it remains27.1% above four weeks earlier.

Still needed: Exact integer daily export and stable-coverage history; Full cache and paid/free splits; realized revenue/compute intensity; Private OpenRouter traffic, direct-provider APIs and first-party apps outside this public series


## CUR-C03: Two distinct $70B references accompany OpenAI’s reported $50B annualized figure.


Axios reported a $50B annualized figure versus an earlier $70B presentation on a different accounting basis. Separately, Bloomberg reported roughly $50B annualized at end-September and an expectation to reach or exceed $70B by year-end, citing people familiar with the matter. The original-page excerpt was retrieved; this verifies the attributed report, not audited financials or a realized same-basis forecast miss as of October 23.

Axios reports $50B versus an earlier $70B presentation on a different partner-sales accounting basis. Separately, Bloomberg’s October 9 original-page excerpt reports roughly $50B annualized revenue at end-September and an expectation to reach or exceed $70B by year-end, citing people familiar with the matter. Neither establishes a realized same-basis forecast miss.

Still needed: Original investor presentation and audited revenue-recognition reconciliation; Underlying company target and accounting comparability across the two reports; Full Bloomberg article not claimed; verified scope is visible original-page excerpt


## CUR-C04: Meta had great earnings; this is bearish for near-term frontier demand.


Meta illustrates monetization inside an established distribution business. The implication for third-party frontier demand is an inference, not a reported result.

Q3 revenue64.601B rose26%; GAAP operating income12.988B fell37% but includes9.89B litigation charge. Ex-charge operating income22.88B. Capex34.91B; FCF−2.12B. Strong core monetization does not disclose external frontier-model purchases.

Still needed: Spend by external lab/model and training/inference; Incremental advertising gross profit causally attributed to specific models; Do not conflate GAAP EPS with charge-adjusted EPS


## CUR-C05: Tesla had great earnings; this is bearish for near-term frontier demand.


Tesla’s quarter is mixed across measures. Its operating results do not measure external frontier-language-model demand.

Q3 revenue28.046B roughly flat YoY; GAAP operating income0.718B down56%; GAAP common net income0.787B down43%; adjusted EPS0.48. Free cash flow1.531B recovered sequentially from−1.092B.

Still needed: Exact meaning of great: consensus beat, sequential recovery, or YoY growth; External model spend; autonomy/training versus general inference split


## CUR-C06: Prepaid, contracted and delayed compute create funding pressure if revenue disappoints.


Separate contract signature, prepayment, delivery, billable service and cash collection. Funding stress is a conditional scenario until contract-level obligations are matched.

CoreWeave’s disclosed backlog definition is conditional on delivery and service availability. The economic mechanism is valid, but the current inquiry has not reconciled customer minimum payments, delivery triggers, unrestricted cash and financing for a named contract.

Still needed: Named contracts, payment schedules, termination rights and delivery remedies; Restricted versus unrestricted funds; Matched supplier/customer financing and revenue cohort


## CUR-C07: Dots, Duet, Grok Bot and Google give away proactive usage, contradicting blanket scarcity.


Some marginal usage is bundled or subsidized. Product terms do not establish unconstrained physical supply; the universal free-agent claim needs these distinctions.

Dots is bundled with eligible paid plans and deeper-work allowances. Duet is a US Pro/Max waitlist preview with background work counting against shared limits. Grok Bot has separate usage bundled in paid plans. Google extends personal intelligence to free-tier users.

Still needed: Measured proactive usage, compute cost and subsidy by plan; Capacity/time/location, latency and limits; No aggregate utilization conclusion from promotional price alone


## CUR-C08: The biggest spenders are shifting to smaller models.


A concrete application vendor reports replacing frontier answering with a specialized model. Broader largest-customer migration remains unmeasured.

Intercom reports moving approximately100% of English chat/email Fin core answering to its own Apex model from frontier models. OpenRouter shows high open-weight token share, but neither establishes migration by a ranked universe of biggest spenders or model parameter sizes.

Still needed: Ranked spender denominator and before/after bills; Comparable task quality and model size; Retained training/distillation/frontier spending


## CUR-C09: Major buyers are waiting six months for cheaper models.


Treat waiting as an explicit scenario assumption, not an observed industry rate.

A bounded search did not locate a current named major-buyer disclosure establishing this exact behavior or its scale. This is not proof that such behavior does not occur.

Still needed: Named buyer/decision-maker statement; Original publication and date; Workload, deferred dollars and actual adoption date


## CUR-C10: Altman says Dot saves him one hour a day.


At13:23, Altman estimates Dot saves him an hour daily on unwanted work.

Original YouTube captions place his estimate at13:23–13:29. Publication metadata is2026-10-21T08:30:24−07:00. This is a personal estimate, not measured cash savings or a universal productivity ceiling.

Still needed: Independent time-use measurement; Configuration and ultrafast status; Whether saved time becomes output, leisure or paid payroll reduction


## CUR-C11: AI coding creates review, rework and bug costs.


Review and rework are real cost categories; measure net delivery time and escaped defects alongside generation speed.

An OSS/Copilot study reports more review/rework concentrated on core developers. METR’s February2026 update warns its later productivity estimates are unreliable because of selection and measurement effects. No current all-coding net-loss estimate follows.

Still needed: Representative current-model controlled workload; Review time, incident costs, maintenance and accepted output; Cash realization versus saved capacity


## CUR-C12: Biotech has brains without legs: intelligence cannot instantly remove lab and trial bottlenecks.


More candidate designs still require experiments and clinical validation before becoming approved treatments.

FDA describes preclinical evidence, human trials and review as distinct steps; Phase2 commonly lasts months to2years and Phase3 one to4years. This supports a physical-validation lag, not a fixed cap on biotech AI demand.

Still needed: Named program bottleneck and compute spend; Laboratory throughput, trial endpoints and approval pathway; Evidence that faster design does or does not expand experimental demand


## CUR-C13: Rising token totals can overstate growth in expensive compute and revenue.


Track token type and realized price: a cached token is not the same economic unit as a newly generated output token.

OpenRouter reports agentic tokens7x human tokens in the last week of September and cached prompts just under90% of all agent tokens. Its classifier and reseller exclusion limit scope; cached tokens, novel inputs and outputs have different economics.

Still needed: Realized provider price and compute intensity by token type; Direct-API and first-party-app coverage


## CUR-C14: A10–20% GDP expansion implies trillions of dollars of additional output.


Using2025 US nominal GDP ofabout30.9T as a scale reference,10–20% isabout3.1–6.2T of additional value added.

Official table gives2025 US nominal GDP of30,861.3B USD.10–20% of that baseline is3,086.13–6,172.26B. These are arithmetic scale equivalents, not forecasts or frontier-model revenue.

Still needed: Causal AI contribution and horizon; Pass-through from value added to enterprise profit and model-provider receipts; Price/inflation treatment for an intertemporal growth projection